NLIDB Progress
Dotting the I and crossing the T in NLIDB. It is fairly time consuming. So many small things that can wreck havoc on your system. Don't feel like working. Too much work for a single person to handle. Will take it slow. Too much to take care of.
my system is pretty good. Captures almost everything nicely. I don't think there can be a better unsupervised NLIDB. maybe a pure semantic parser which eliminates the pipeline but I doubt it will perform better than my system.The level of detail captured by syntaxnet is quite good. These details help a lot in the final result.
The code has become bloated. I need to trim it.
I think I will add natural language searches and boolean searches in database as back up methods in case I get poor scores from scoring algo. Ecommerce sites are going to have a lot of grammatically incomplete sentences. Stringing the columns together is the way to go.
Basic version is more or less done. Finalizing it is too much work.Spelling errors are going to be a pain in the ass. Norvig's spelling handler is very basic. Plus, I will need to get a corpus. Plus domain specific handling is going to be an issue. I need a better model.
my system is pretty good. Captures almost everything nicely. I don't think there can be a better unsupervised NLIDB. maybe a pure semantic parser which eliminates the pipeline but I doubt it will perform better than my system.The level of detail captured by syntaxnet is quite good. These details help a lot in the final result.
The code has become bloated. I need to trim it.
I think I will add natural language searches and boolean searches in database as back up methods in case I get poor scores from scoring algo. Ecommerce sites are going to have a lot of grammatically incomplete sentences. Stringing the columns together is the way to go.
Basic version is more or less done. Finalizing it is too much work.Spelling errors are going to be a pain in the ass. Norvig's spelling handler is very basic. Plus, I will need to get a corpus. Plus domain specific handling is going to be an issue. I need a better model.
Comments
Post a Comment